Corpus Linguistics in the South #11 by John Williams

Perez-Paredes: representation of immigrants in UK admin

johnxwilliams1 — 2016-02-27 10:39:23 UTC

LADEX corpora - comparing representation of immigrants across different languages and countries --> consist of admin texts, laws, etc

UK media portrayal of migrants tends to be more negative than in the Spanish media

The word 'immigrant' does not occur in UK law

Corpus-driven study with both qualitative and quantitative aspects

Collocates and Word Sketches of 'migrants' were collected

Findings from UK corpus (about 5 million?):

Overtly negative representations were avoided.
Immigrants are constructed as a largely homogeneous group.
Top collocate is 'illegal'
No foregrounded nationalities
'Partially constructed individuals' - associated with very few semantic categories
In the law corpus, 'migrant' is associated with very few verbs

How do we know if our corpus is representative ?

Oakley: Census Corpus

johnxwilliams1 — 2016-02-27 11:17:00 UTC

Census corpus = a corpus which comprises (nearly) all the texts in a given population

The example presented was a corpus of sex education books for teenagers - c. 93,000 words

Some problems of representativity disappear - but when does this stop being CL?

Oakley argues this is CL: it allows visibility of features not visible to the naked eye, eg. diachronic change in the use of the word 'help' in relation to gay sexuality

Bowie & Wallis: Changes in structures and collocation, from treebank to megacorpus

johnxwilliams1 — 2016-02-27 11:48:14 UTC

Investigating declining use of perfect infinitive, to have +ppt : the only way to express pastness in an infinitival clause

For AmE, Used COHA, counted instances & compared against other past forms --> massive decline of around 80% in the period covered by COHA

For BrE, used ICE-GB, used tree-parse to search for form (Fuzzy Tree Parse) : in 84% of cases, structure is complement of verb, 78% of these are in active simple pattern (eg. claims to have). Top three verbs are seem, appear, say. Semantic groups are 'seeming', 'cognition & saying', 'modality', 'prospective meaning' (eg. hope, expect)

The verbs have declined at different rates.

'Prospective' group has declined when it expresses counterfactual meaning (I hoped to have seen him), + remember to have structure.

No other structure obviously replacing perfect infinitive.

Grammatical changes tends to be unevenly distributed lexically.

Turner: Getting to grips with a megacorpus

johnxwilliams1 — 2016-02-27 12:25:04 UTC

Using the Oxford English Corpus to study some and any

Looked for face-saving/threatening uses in OEC and BNC (Do you want some/any ?) Despite differing size, BNC has a better selection of examples because of spoken component?

OEC better for wh- patterns, eg. "What is life without some fun?". Complex pragmatics around these patter

Used the Legal English subcorpus to investigate variation across different legal genres. However, you need to know the quirks of the composition of the corpora in order to draw general conclusions about the genre.

Some evidence for any having negative semantic prosody, more in set phrases than general collocates.

Ozon & Green: Frequency & grammaticalization in Cameroon Pidgin English (CPE)

johnxwilliams1 — 2016-02-27 12:49:15 UTC

Spoken by 50% of Cameroon population (= 22 million), especially urban, only spoken, stigmatized; no verb inflections

Corpus about 240,000 words, partially POS-tagged

High frequency of five 'light' verbs: make, get, take, give, do

In general, light verbs tend to resist grammaticalizations, eg. contractions, but in CPE there is more evidence for this...

...most strikingly, make - mainly grammatical (causative or deontic modality), do - mainly lexical

na (= verb be) - transitions from copula, to cleft marker, to focus marker

Go as a future marker.

Evidence for give having a *beneficial* function (cf. for)

Louw: Small Corpora & The Next Quark

johnxwilliams1 — 2016-02-27 15:16:42 UTC

Took a broader view of the philosophical implications of CL

Methodology = 'after path'

William Whewell, Theory of Scientific Method --> came up with the terms colligation and consilience

When two words, previously unconnected, start appearing as a collocation (eg. truth & reconciliation), something ideological might be going on

"Once in every seven years there is a 'mercenary' event." --> ie 'mercenary' spikes in corpora

The most frequent lexical instantiations of a grammar pattern allow you to read the hapaxes in detail.

Asked audience to check which nouns are used most frequently in the pattern 'is its * with' --> tends to be the semantic area of relationship or connection

Note to self

johnxwilliams1 — 2016-02-27 15:49:58 UTC

Mike Scott's Lexis/Nexis downloader

Collard: Nanocorpora for interpreting

johnxwilliams1 — 2016-03-01 16:28:49 UTC

Normally, difficult to access corpora of interpreted text and transcription is time-consuming.

Windfall! Access has become available to interpreted text from the EU parliament.

Working corpus of < 250,000 words of English, interpreted from French and Spanish

Case study: The use of 'well' by interpreters.

'Well' not always triggered by the source text.

Interpreters tend to both omit and add more connective markers than translators --> imitating spontaneous speech ?

Interpreters hedge more than original speakers, and female interpreters more than male.

Interpreters produce fewer face-threatening utterances than original speakers, and females fewer than males.

(Tentative finding) Female interpreters can keep concepts in their memory longer than males.

Programme

johnxwilliams1 — 2016-03-01 19:06:38 UTC

Viggiano & Williams

johnxwilliams1 — 2016-03-01 19:24:03 UTC

Screencast of corpus-building. Click on link then 'View Original'

Viggiano & Williams: Capturing online forum data

johnxwilliams1 — 2016-03-01 19:31:30 UTC

Slides